Spencer Dawkins a écrit :
Agreeing with Iljitch's math here, but ...
In case people aren't already aware of it, I think the work at the
second URL below is very interesting in discussing the possible
implications of CGN, and in particular the number of outbound and
inbound ports needed per active host.
Hm, it looks like the high water marks for TCP and UDP sessions per
subscriber (including non-active subscribers) is around 20 and 40.
For our purposes we can probably assume that the UDP is lower than TCP
because UDP in this is very likely mostly DNS which shouldn't have to
go through a NAT64.
So that means a single IPv4 address can support about 3000 subscribers.
... the "planning for 500 connections per user" reference in the plenary
presentation last week was way higher than I've heard previously, but
the working number from the P2PI workshop in May was about 80 for
BitTorrent,
I think Miyakawa-san meant "up to 500", and he was referring to AJAX
connections used in applications such as google maps.
if the deployment scenario is based on multiple computers behind a NAT
in the home, and 200 connections per computer and a few people in the
home accessing Internet services, given that even now the TiVO and other
boxes are accessing Youtube and other services, then I think 500 is a
good cap number since it will happen over time. We shall need to take
this number as a reference point for our work.
Marc.
wasn't it? So we're already looking at a factor-of-two
reduction in what we can support on one IPv4 address, before someone
comes up with an even cuter connection-use strategy...
:-(
Spencer
_______________________________________________
Int-area mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/int-area
_______________________________________________
Int-area mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/int-area
